Fri 22 Jun 2018 16:10 - 16:35 at Grand Ballroom AB - Parallelism Chair(s): Julian Dolby

In this paper, we develop an approach to GPU kernel optimization by focusing on identification of bottleneck resources and determining optimization parameters that can alleviate the bottleneck. Performance modeling for GPUs is done by abstract kernel emulation along with latency/gap modeling of resources. Sensitivity analysis with respect to resource latency/gap parameters is used to predict the bottleneck resource for a given kernel's execution. The utility of the bottleneck analysis is demonstrated in two contexts: 1) Coupling the new bottleneck-driven optimization strategy with the OpenTuner auto-tuner: experimental results on all kernels from the Rodinia suite and GPU tensor contraction kernels from the NWChem computational chemistry suite demonstrate effectiveness. 2) Manual code optimization: two case studies illustrate the use of the bottleneck analysis to iteratively improve the performance of code from state-of-the-art domain-specific code generators.

Fri 22 Jun

pldi-2018-papers
16:10 - 17:25: PLDI Research Papers - Parallelism at Grand Ballroom AB
Chair(s): Julian DolbyIBM Thomas J. Watson Research Center
pldi-2018-papers16:10 - 16:35
Talk
Changwan Hong, Aravind Sukumaran-RajamOhio State University, USA, Jinsung KimOhio State University, USA, Prashant Singh Rawat, Sriram KrishnamoorthyPacific Northwest National Laboratories, Louis-Noel PouchetColorado State University, Fabrice RastelloINRIA, P. SadayappanOhio State University
Media Attached
pldi-2018-papers16:35 - 17:00
Talk
Roshan DathathriUniversity of Texas at Austin, USA, Gurbinder GillUniversity of Texas at Austin, USA, Loc HoangUniversity of Texas at Austin, USA, Hoang-Vu DangUniversity of Illinois at Urbana-Champaign, USA, Alex BrooksUniversity of Illinois at Urbana-Champaign, USA, Nikoli DrydenUniversity of Illinois at Urbana-Champaign, USA, Marc SnirUIUC, Keshav PingaliUniversity of Texas at Austin, USA
Media Attached
pldi-2018-papers17:00 - 17:25
Talk
Umut AcarCarnegie Mellon University, Arthur CharguéraudInria, Adrien Guatto, Mike Rainey, Filip SieczkowskiUniversity of Wrocław
Media Attached